The Should-I-Split Check

Should this be one agent or several?

The task: [DESCRIBE]
Current design: [DESCRIBE]
Why I am considering splitting: [DESCRIBE]

Assess honestly:
1. Are the sub-tasks genuinely independent, or sequentially
   dependent?
2. Does each sub-agent need context the others have?
3. What is the coordination overhead — briefing, waiting, reading
   reports, reconciling?
4. Would a single agent with a longer run be simpler and cheaper?
5. Is the split solving a real problem, or is it architecture
   theatre?

Multi-agent is genuinely justified when:
- Sub-tasks are independent and parallelisable
- Different sub-tasks need genuinely different tool sets or
  permissions
- Context isolation is a requirement, not an accident
- One sub-task would exhaust context on its own

Multi-agent is NOT justified by:
- "It feels more organised"
- Sequential steps a single agent could do in order
- Wanting different prompts — that is a prompt problem
- Wanting a verifier — often better done in the same loop

Give me the honest recommendation, and if you recommend splitting,
the specific cost in extra tokens and latency.